mastodon.social is one of the many independent Mastodon servers you can use to participate in the fediverse.
The original server operated by the Mastodon gGmbH non-profit

Administered by:

Server stats:

380K
active users

Jesse

People like to say that "AI" is different from crypto in that there are actual useful applications, and that's true. But the vast majority of people you're expecting to come up with those applications are the same people who were just trying to build products on the blockchain.

I kind of understood this, but it's getting clearer to me. The kind of people who build good thoughtful products are going to be disproportionately put off by AI and the surrounding culture, leaving the field largely to scammers and magical thinkers. And the ensuing flood of shit isn't neutral. Actual good stuff gets lost. I really think this kind of hype is bad for real innovation.

@misc The prevailing wisdom on crypto always seemed backward to me: Many people said that yes Bitcoin et al. were silly but the “real” and “serious” applications were in the underlying blockchain tech. In reality, Bitcoin is as useful as the blockchain gets (not very lol), way more practical than using a blockchain for logistics or whatever. “AI” could be similar—maybe ChatGPT is as good as it gets.

@eARCwelder

I've said before on this federated microbloggy network thingy

I'm maaaad that the most _fun_ application of LLMs — zany weird lateral thinking tools like a tarot deck or Oblique Strategies on steroids — has been torn from my hands and replaced with something that carefully generates banal spongecake content guaranteed not to surprise anyone

trochee.net/samyro/

@misc

Trochaisms - Letters make words; sentences make paragraphs · SamyroSamyro what ipsum beast lurches towards Bethlehem to be born? An automatic poetry organizer built on TensorFlow’s RNNs.  In Python, using the C bindings built into TensorFlow for speed. Samyr…

@trochee @eARCwelder Well, we can still use the old stuff right?

I do feel like that use case for the artistic process, making weird stuff to inspire yourself with, does still seem possible with the current gen of image generators, more so than with the text stuff.

@misc

The modern crop is all built on stolen valor ("foundation models" that have scraped the whole Internet)

It's hard to match their slickness with the older, honest stuff
@eARCwelder

@trochee @eARCwelder Yeah no I get that but also I think it can be plausibly acceptable to use it as part of an internal process. Like, a longtime fantasy of mine is just having a big personal database of film stills. It feels a little like an elaboration on that. To be able to zero in on a look or a vibe. I wish the provenance was traceable though (massive understatement)

@misc

i agree with you. I'm worried about the flood of garbage coming.

@jmcrookston @misc

Get ready for automated police and hiring racism, even more arbitrary health insurance and loan denials, and the general return of social Darwinism.

That's the major thing I see LLMs contributing, along with generally making nearly all written material useless filler garbage.

@eestileib @misc

Yup the inextricable racism.

Also it'll give a lot of cover to deliberate racism; that's my guess.

@misc Totally agree. As if the web wasn't completely kaputt before, now we're being swamped with generated zero-effort, zero-use nonsense all the time.

I'm really thinking hard about the consequences and one of them is that I think the web as we know it really will dwindle in the coming years.

Many knowledgeable people have suggested that blogging will come back, and I kind of hope so, but I think we really don't know what human nature will prefer in terms of *actual* authenticity.

@ftranschel @misc
In an age of machine-generated SEO spam, web portals and directories are due for a comeback.

Soon: Join Unikitty's webring of artisan human-created content!

@unikitty @misc And behold, Pagerank enabled Web 2.0 and it also taketh it away.

@ftranschel @unikitty @misc

AdSense took it away. Google didn't have to start linking only to pages built to serve the ads from another division of the conglomerate. They killed their own golden goose.

@ftranschel I'm more optimistic given how much people seem to want to return to a more small but genuine internet of the early 00s. The internet as we know it might not survive but that aspect, the need for people to share themselves to the world, that won't go away.

@nini Sure, for you and me that might very well be the case. But Instagrammers and TikTokers out there, I'm *not* optimistic...

@misc

It's truly amazing to see overhyped tech from a decade ago turn up unheralded in, like, sex toys.

@Dseitz lol wait you can't just leave that there without saying more

@misc

I did recently see a voice activated dildo in passing on some social feed.

@misc I think this is true of chatbot style models, their equivalent for visual and sound outputs and anything that is "generalised". However the really niche stuff like protein folding is different despite using fundamentally the same techniques. I do suspect that the cross over in workforce is probably very small though.

@misc Yes. Forbes has predicted a financial crash caused by the adoption of AI (not explicitly AI itself) within 10 years. I've warned work of several exploitable vectors if used for resume/CV evaluation in HR. (Guarantee your resume is #1, always, for any job.)

I appear to be alone at my corporation in being weary of the technology rather than diving head-first. Slapped a co-worker for asking ChatGPT about a specific solution instead of Google about the problem. Wrong answer v. first result.

@alice @misc I use both Google and Chatgpt, but I would honestly not need chatgpt if the Google search engine was as good as it was 5-10 years ago. The alternatives are still not at that level unfortunately either in my experience

@xWood4000 @misc Well, I used "Google" in the generalized dictionary sense. I personally use DDG, though did develop some concern over their reversal on the "not filtering your results" aspect. Regardless of the particular reason.

I like stable result ordering. Makes result sets sharable, or, at least, simultaneously experienceable.

@misc

This is beautifully said. I have been thinking similarly about writing since AI came on to the scene, but you have put it more succinctly.

To borrow from your quote, “The kind of people who are good writers are going to be disproportionately put off by AI and the surrounding culture, leaving the field largely to scammers and magical thinkers.”

@misc

As a designer it's rocket fuel.
It takes maybe a month to design and build a 3d object fully..with full exploration..which is what takes time.

With ai I can do the base design and then quickly get 100 variations and pick the best bits of each. repeat..

The result will be more advanced because you have effectively iterated more. Design quality largely comes down to the amount of iterations.

The designer could do the same thing, mostly, but time is an enemy.

Design will improve.

@LewisHarrington I'm curious what tools you're using... I don't know much about 3D modeling. Do you mean like using image generators for inspiration, or are there purpose built features in your modeling tools, analogous to something like Gitlab Copilot?

@misc I use 2d drawing progams like Affinity designer.
This program uses quite simple shapes though.
The results are variable. It's a question of experimenting with the word prompts and using the generated images to photobash to create a design.

krea.ai/home

Once the 2d aspect is done I will build the output as a 3d model. I use Plasticity mostly to make the model.
I just found Krea a few days ago so it's all new really.

plasticity.xyz/

@misc oh yeah this is a point I made at the end of my recent paper -- "working in AI" has a field-specific culture itself and that culture has big problems. This isn't a new point of course re: diversity and exclusion in the field but the way it permeates into newly-formed societal field-specific ability beliefs is an area worth of study and scrutiny imo. We keep DOING this (same evolution in many parts of data science) with reliable bad outcomes

@misc

This is a really good insight. I had not thought of "AI" that way before, but you're right.

@misc Idk, there's a number of next gen AI products that I think are genuinely useful and quietly succeeding: GitHub Copilot works great for me. There's a lot of creative professional software that's incorporating modern ML models in ways that seem genuinely useful (my partner works on audio editing tools and their AI stuff seems along similar lines of usefulness to autotune).

Of course, I'm talking purely about usefulness here, ethical questions about training notwithstanding.

@harris I don't disagree - but there is *so much shit*. I keep thinking about the iOS App Store, which I followed obsessively for years. (undiagnosed OCD, yay!) I saw so many really cool apps fail to get traction, and *so many* flappy bird clones (just to name one example). The gold rush hype inhibited the emergence and success of good products.

@harris And the underlying atmosphere is even worse here, especially given the reserve army of blockchain hucksters who are already flooding in. I'm not surprised that real innovations would be more likely to come from established organizations with their own cultures and varying degrees of insulation. (Apple being the obvious example)

@misc
There are also the rest of us who have been using the "new-found" NN and embedding models in statistics for our whole careers.
We didn't pivot when the wave washed in, and we'll still be here when it washes back out.
It's amazing how little our work has changed. We have the same problems working about as well. We have the same glacial rate of progress.
Perhaps we'll benefit by the surplus of cheap cloud GPU leftover? It's too bad all the ASIC hardware left by miners is useless

@misc Here is a tip: products that use AI for marketing are garbage; products describing their functionality as machine learning/ML have actual use.

@misc deepl.com still going strong since 2017. By today's definition, they are doing AI.

@misc sure hope that the next buzzword they used to defraud venture capitalists isn't one that fucks up the environment or fucks over the lives of workers

@misc It’s such a disingenuous argument, because we’ve already been using these same general algorithms for *ages* with good, clean data, in the targeted scenarios for which they’re actually suited.

The only major difference nowadays is we have access to enough compute and storage to dump the Internet’s trash bags into them and still get something that resembles coherency, and we call it a miracle. ¯\_(ツ)_/¯

@misc AI is like a pitbull. While some owners might do a better job than others with training, there is a potential for catastrophe unique to the species.

@misc Pretty easy to disprove your message. None of the established tech companies got into the crypto, blockchain hype. They had the resources, but they didn't. They are investing like crazy on AI. I have heard a lot of respected technical people get on board. Scams and hypes are an everyday thing for Wall Street. Just because someone is scamming, doesn't mean there isn't an actual good product elsewhere.

@notabird I'm not sure what point you think you just disproved but I don't think it was mine?

@notabird @misc: "None of the established tech companies got into the crypto, blockchain hype."

Several of them did. IBM, Facebook, Microsoft and others openly did; other big companies likely had skunk works projects in the background. It's just that "AI" is an established term and is easier to sell from a marketing perspective than "internet funny money made up by a bunch of ancap weirdos".

@misc The trend chasers do exist, this is true, but mostly on the business side.

On the technical side, however, there are tons of developers who never touched crypto, because they saw it for the scam it was, who are now excited about the possibilities of AI.

This is normal, though. The idiots with the money will always flit around chasing the buzzwords while the rest of us slowly grind the gears to turn it into something useful.

@misc yeah, there are useful applications alright. nothing they're selling you is a useful application, but they are out there

@misc Salesmen (tech bros) who lead their pitch with the technology they are using should not be trusted. AI isn't magic pixie dust that will build useful applications all by themselves - nor will it have clever ideas for which applications to build. If the salesman has something concrete to sell, he will lead with the application. If he wants you to fund his brainstorming, throwing stuff against the wall exercise, he will lead with the technology.

@misc What?! Does AI mean the blockchain hype is now over?!

I guess every cloud (pun intended) has a silver lining. 🤪

@misc these backend compute servers excel at using fast amounts of electric power as well.

@misc @jcrabapple eh, i’m not a fan of criticizing people instead of ideas. that’s a big red flag for me. also, in my experience, i rub shoulders with a lot of people working on AI applications, and none of them (that i’m aware of) worked on blockchain tech 🤷‍♂️

@kellogh @jcrabapple I'm not sure I follow what the red flag is? Not criticizing anyone in particular, just generalizing from what I've observed. I'm sure your colleagues are great and doing cool work.

@misc when an argument moves past the merits alone and instead focuses on people and who they are, it’s a good sign the productive parts of the conversation are over. i can’t think of many exceptions to that. generalizing about people is never good, as a rule. it certainly doesn’t make things better

@misc There may be useful applications but I think the cost to the environment outweighs the benefits. (Specifically for the currently hyped "AI"; it is possible to do frugal machine learning and I have no issue with that.)